Automatic detection of ship target and motion direction in visual images

摘要

In this article, we focus on automatic ship target detection in visual images. The proposed approach consists of two stages. At the first stage, emphasis is placed on sea background suppression. A pre-processing technique is studied based on mean-shift smoothing algorithm. In this process, the improvement is that we associate the range bandwidth of mean-shift with local image properties, thus achieve the purpose of suppressing sea background while protecting ship target. At the second stage, a hierarchical ship target detection approach is developed. To begin with, region-of-interest (ROI) of ship target is coarsely identified via salient edge regions extraction. Then main body of ship target is located by projection, prior spatial structure information of ship target is used for further ROI refinement. After that, a post-processing procedure based on colour features is presented to obtain final detection results. One advantage of the proposed method is that motion direction of the ship target can also be estimated, which could provide important information for ship target tracking and monitoring. Tested on real visual ship target images supplied by the co-operator, experimental results show that the proposed algorithm can effectively detect big ship targets and its motion direction with various backgrounds in visual images.